Validation of a diffeomorphic registration algorithm using true deformation computed from thin plate spline...

Area of Science:

• Medical Imaging
• Computational Anatomy
• Biomedical Engineering

Background:

• Manual contours are frequently used for validating nonrigid medical image registration, despite inherent variability.
• Accurate validation methods are crucial for advancing medical image analysis and segmentation techniques.

Purpose of the Study:

• To introduce and validate a novel method for creating a true deformation field using thin plate spline interpolation.
• To compare the performance of a diffeomorphic registration method against other algorithms using simulated cardiac motion data.

Main Methods:

• Thin plate spline interpolation was used to generate a ground truth deformation field.
• A diffeomorphic registration method was evaluated against three other algorithms (RealTITracker L2L2, RealTITracker L2L1, Elastix).
• Performance was assessed using Dice scores on simulated cardiac motion deformation from 10 subjects in the Automated Cardiac Diagnosis Challenge (ACDC) dataset.

Main Results:

• The diffeomorphic registration method achieved superior performance with Dice scores of 0.991 and 0.997.
• Compared to other methods, the diffeomorphic approach showed more robust and accurate results in cardiac motion registration.
• Elastix also demonstrated strong performance, particularly in the second registration approach.

Conclusions:

• The proposed thin plate spline interpolation method provides a reliable ground truth for validating registration algorithms.
• The diffeomorphic registration method exhibits robust performance and is suitable for accurate cardiac motion analysis and segmentation.
• This validated registration method has significant clinical relevance for the segmentation of heart chambers.